This repository is designed to run inside a Docker container. It requires Docker Engine and nvidia-container-toolkit as prerequisites for building the container or pulling it from Docker Hub. Follow the instructions below to install the required dependencies.
After this, the docker image can be built by executing following command inside this github repo folder Scrubland-Field-Delineation
sudo docker build --progress=plain -t <image-name> .
Use the following command to disable progress display.
sudo docker build -t <image-name> .
Instructions to pull docker image comming soon...
To start docker container please run the following command.
sudo docker run --shm-size=60gb --gpus all --init -it -v $(pwd):/app <image-name> bash
Inside the container, the required environment is pre-activated for convenience. The script.py is the one that needs to be executed to obtaine vector boundaries of fields, plantation and scrubland. Before running the script, download the model (india_Airbus_SPOT_model.params) from here and place it inside Scrubland-Field-Delineation folder. After this you can run the script using the following command.
python script.py
When a new container is started for the first time, this will prompt a link for authenticating into Google Earth Engine. Click the link or copy-paste it into a browser, then log in using a Google account associated with Google Earth Engine. After authentication, you will receive an access key, which should be pasted into the Docker container to complete the authentication process. Once authenticated, the script will process the specified Region of Interest(ROI) and generate vector boundaries for farms, plantations, and scrubland.
You can set the ROI inside the script.py after ee.Initialize in __main__ block. A boilerplate code for running on a rectangular region instead of any polygon ROI is also provided after the ROI declaration which commented out for ease of testing. Also set the name of the directory which will be used to download images, store predictions and store vector boundaries.
| Module | Description | Link |
|---|---|---|
| Local Compute Script 1 | Implements AEZ tiling, sampling, and boundary generation | View |
| Local Compute Script 2 | Applies rule filters and generates stratified samples | View |
| GEE Script 3 | Temporal multi-year sampling and classifier training | View |
| Farm–Plantation Boundary Classification | Cleans and exports farm and plantation polygons | View |
| Boundary Generation (Block Level) | Block-level tile generation for ROI-based runs | View |
| Adding new grids | Here we add new grids to generate data from those grids to improve the lulc further | view |